Financial News API: Extract Earnings Results Without Mixing in Guidance

An earnings article can contain several revenue figures that answer different questions. One describes a completed quarter. Another is management’s estimate for the next quarter. If a financial news API workflow drops those labels, a dashboard can display a forecast as reported revenue.

Reddit’s October 29, 2026 earnings release provides a current example. The company announced results for the quarter ended September 30 and separately gave fourth-quarter estimates. Both appear in the same document, so identifying the company and extracting a dollar amount is insufficient.

For developers building earnings alerts or research assistants, the useful output is a claim with its period, status and source. The workflow below describes application logic for producing that output, rather than a claim that an article feed supplies a finished financial database.

Classify each claim before extracting its value

Give reported results, company guidance and outside estimates separate categories. Preserve the speaker or organization responsible for each claim. A journalist quoting an analyst’s forecast has not turned that forecast into company guidance.

Use the surrounding sentence and section heading to establish the category. If the supplied text is too short to determine it, mark the claim for review. Do not infer that the first revenue number in an article must describe the latest completed quarter.

In Reddit’s release, the distinction is explicit. The table below summarizes two revenue claims from that source; it does not estimate future performance.

Keep a range as two bounds with its original label. A display may need a single value for a particular calculation, but any derived midpoint should remain separate from the company’s stated range.

Store the period with the metric

Record the fiscal period as stated, including the year. Keep the publication date in another field. An announcement made in October can report activity from a quarter that ended in September.

Do not assume that every company’s fiscal quarters follow the calendar. Use the period identified in the source and retain an end date when supplied. Where the article only says “last quarter,” obtain supporting context before assigning a period.

Store the comparison period separately too. A year-over-year change and a quarter-over-quarter change measure different intervals. If an application compares them as equivalent growth figures, a correctly extracted percentage can still produce a misleading result.

For example, a synthetic test sentence could read: “Revenue rose 8% from the previous quarter.” Its expected comparison is quarter-over-quarter. A test should fail if the extractor labels it year-over-year, even when the numeric value remains correct.

Preserve units and accounting labels

Keep the currency, scale and metric name together. “895 million” and “895 billion” are different observations. Record the original presentation alongside any normalized numeric value so a reviewer can inspect the conversion.

Preserve qualifiers such as adjusted, diluted and constant currency when they occur. Do not merge differently labelled measures because they share a company, period or similar wording.

Reddit reports adjusted EBITDA alongside other financial results and provides a section explaining its non-GAAP measures. That is a reason to retain the precise metric label and supporting reference, rather than shorten every earnings measure to “profit.”

Route an ambiguous label to review. If the source contains only “earnings improved,” do not manufacture a net-income value or accounting basis. An empty field with an explanation is more useful than a plausible number that the article never supplied.

Retain enough source text to review a claim

Attach the source URL and a short supporting passage to each extracted claim. Include the relevant heading when it establishes that a value is guidance. A numeric sentence detached from its section can lose the distinction the pipeline just worked to preserve.

Keep the source’s publication time and your application’s processing time separate. For corrections, retain the earlier observation and identify which source version supports the replacement. Processing an old release today does not make its figures newly announced.

Where multiple articles quote the same release, connect those claims to their common origin where established. Repetition can help locate coverage, but it does not provide several independent measurements of the company’s revenue.

Give the news feed and extraction layer different jobs

Webz.io’s News API provides article text, metadata and enrichments in structured JSON, including entities and sentiment. Its product page also describes content from corporate and government newsrooms. These features support retrieval and company monitoring.

Your application still needs to determine the financial claim’s type, period, units and label. Check the fields actually returned by your integration before designing a schema around them. The proposed earnings fields here are application requirements, not documented API field names.

Use Webz.io’s financial news API overview for the broader collection use case. For this narrower workflow, evaluate whether the retrieved text contains enough context to distinguish a result from an estimate. A relevant headline alone may not.

Test the distinctions that change the output

Create a small evaluation set before enabling alerts. Include one document containing both results and guidance, one with a forecast range, one with two currencies and one discussing a previous year’s quarter.

Add deliberately incomplete snippets. The expected response should be a review state when a period or claim type cannot be established. Measure those decisions separately from successful numeric extraction.

Check the final alert as well as the stored record. A fictional alert could read: “Example Company; Q4 2026 revenue guidance; USD 100–120 million; company estimate; source reference E-7.” This tests whether the display preserves the range and attribution.

Have a reviewer compare the output with the supporting passage. Record errors by cause: wrong period, lost qualifier, unit conversion or claim-type confusion. Use those categories to choose the next test cases.

Correct the claims already delivered

When a source correction changes a figure or label, identify the affected claim and downstream alerts. Retain the earlier value with its status, then publish the corrected observation with its supporting reference.

Test that process using a synthetic correction before relying on it. Confirm that the dashboard changes, the previous alert remains traceable and the correction reaches the person responsible for reviewing it.